The Energetic Beat: Exploring the Relationship between Tempo and Danceability in Hip Hop Music

In general, hip hop music is known for having a strong beat and a high level of energy. This is reflected in the tempo and danceability of many hip hop tracks, which are often fast and rhythmically complex.
This plot really shows this relationship. A large proportion of tracks correspond to high energy and the “danceability” is well over 0.5 on average. Furthermore, we see interesting differences such as that the average tempo of Dutch raps is lower than French raps. The tempo distribution of French raps is greater than that of Dutch raps. This may be the reason why French raps more often give me more energy than Dutch raps.

Rhythmic Resonance bewteen Valence, Energy, and Loudness in French and Dutch Raps

This plot shows the relationship between the features “valence”, “energy” and “Loudness” in French and Dutch raps. In the figure, the higher the valence and tempo be, the higher the loudness of the tracks be. Furthermore, the average energy of Dutch raps is lower, but whether this is significantly lower is questionable, as a discussion point is that the corpus contains slightly less Dutch rap than French rap. Lastly, the distribution of valence is almost the same in both cases

Histograms of key modes for French and Dutch tracks in the corpus (WEEK 10)

This graph shows the number of times a major/minor key appeared in the entire corpus. It shows that almost every key appears in the corpus, but looking at the Dutch corpus, the D# major does not appear and F major appeared only once. This result corresponds to the French corpus in which D# major appeared only once and F# major did not appear at all. Moreover, A# major did not occur at all in the French corpus while it occurred five times in the Dutch corpus. Among other things, this would perhaps explain the difference between the two corpora. In the end, no difference is directly visible in the number of times a particular key did or did not occur in the corpus.

SD Tempo vs. Mean Tempo (Track-Level).

The standard deviation, denoted by σ, is a metric that quantifies the degree of dispersion of data with respect to the mean. When the standard deviation is low, the data points tend to be tightly clustered around the mean, while a high standard deviation implies that the data points are widely spread out [1]. In this graph, for each track, the Standard Deviation of “tempo” is plotted against the Average Tempo (bpm) in that same track. This has been done for both French and Dutch Raps. Thereby, the variable Volume (in dBFS) and Duration (in minutes) can be read from the graph. From the graph we can see that, as shown earlier, Dutch raps are clustered closer together when it comes to tempo. In this graph, you can see a similar result, namely, that Dutch rap has a lower average tempo than French Raps. I experience this difference in tempo myself when listening to Dutch and French raps. The average tempo of French raps are generally more spread out. The Standard Deviation in tempo for both genres (in this case languages) do not differ much from each other. We can see that when the average tempo is lower, the Standard Deviation is higher. Finally, we also see that tracks with a higher volume have a higher average tempo and a lower Standard Deviation and generally last more than three minutes.

Reference: 1. U.S. National Library of Medicine. (n.d.). Standard Deviation. https://www.nlm.nih.gov/nichsr/stats_tutorial/section2/mod8_sd.html

The most promising herald, poised to set these languages asunder, an unmistakable marker of distinction! (Track-Level).

In this graph, the average timbre coefficients in French and Dutch raps were analysed to come to a conclusion about which coefficient might make the most difference. Looking at the graph, we see that almost all coefficients are the same for both genres. To still point out a difference, we can say that coefficient 7 differs the most. So, coefficient 7 looks like the most promising marker distinguishing these genres, but we should verify that with cepstrograms and listening tests of specific pieces.

Tempo Tales: Exploring the Rhythmic Differences between French and Dutch Tracks

The histogram shows the overall estimated tempo of each track in both corpora (French and Dutch) in Beats Per Minute (BPM). In musical terminology, tempo is the speed or pace of a given piece and derives directly from the average beat duration [1]. If we look at the density plots, we can see that the tempo of Dutch tracks has been distributed differently from those of French tracks. As it happens, Dutch tracks mainly have a tempo between 90 and 105 BPM, whereas the tempo of French tracks has been more equally distributed between 90 and 150 BPM. These results could also be seen in the “The Energetic Beat” plot from week 7 in “Feature analysis”, where the relationship between Tempo, Danceability has been explored.

1. Web API Reference | Spotify for Developers. (n.d.). https://developer.spotify.com/documentation/web-api/reference/#/operations/get-audio-features

Tempograms of a recording of PNL; Au DD” (French)

The following plot is a Tempogram for “Au DD”. As it happens, the track has a relatively steady tempo with occasional variations. I think taht these variations (highlights) are visual noise which is a product of tempo octaves. From the 150th seconds onwards, you can see that the relatively stable pattern changes to a higher tempo (between 150th and 170th second) and suddenly to a lower tempo (around 175th second). This can be attributed to the fact that between 150th and 170th second there is more music dropping out paired with a higher instrumentalness and speechiness. In the 175th second, there is a short silence both in speechiness and music overall. Furthermore, at the end, it can be seen that there is more frequent tempo changes, which can be explained by the sluggish instrument playing (usually every track ends like this). In the tempogram of this track the highlights at the end, seen in it’s self similarity matrices are visible.

Tempograms of a recording of Boef; Treinstation” (Dutch)

As compared to the previous track “Au DD”, this track “Treinstation” exhibits more frequent tempo changes and has more highlights besides its relatively stable tempo curve. from the moment the 30 tyh second is reached, the tempo curve leaves its stability for a short time (around 30 seconds) due to the change in instruments. This also happens in the 80 th second where more instruments drop out. In the tempogram of this track, the change in tempo curve could also be seen in the self similarity matrices, especially at points where some pieces get repeated (parallel lines to the diagonal).

Hierarchical clustering

There are three primary types of linkage: single, average, and complete. In this dendrogram, the type “complete” was chosen because it gave the best cluster result. Using complete clustering for a playlist of fifteen French and fifteen Dutch tracks, one can clearly see that there are two main clusters. The reason that there are not of all tracks (there are 337) is because it became computationally very heavy and the graphs would otherwise look very cluttered. The left cluster clearly contains (10/12) x 100% = 83.3% Dutch raps and the right cluster consists of (13/18) x 100% = 72.2% French raps. Thus, the language and features of the raps are clearly visible in this clustering. From this, it can be concluded that French and Dutch raps clearly differ from each other. However, it remains to be seen which features cause this difference. For this we will have to see which features are most important in creating clusters. This can be seen under “Feature Importance (Feature Set)”.

Heatmap

Looking closely at this heatmap, one can see that the bottom left part of the heatmap is clearly lighter in colour. In addition, the top right part is also clearly lighter (greener/yellower) coloured and the middle forms a dark coloured diagonal from top left to bottom right. What can be concluded from this is that most Dutch raps (bottom left) clearly differ from French raps (top right) when it comes to features c08, speechiness, c06, C#/Db. Indeed, these features are stronger in Dutch raps than French raps. Moreover, we see that French raps clearly differ from Dutch raps when it comes to c09, instrumentalness, c12, and tempo. Earlier, we also saw that tempo was clearly a features that was less widespread in Dutch raps and more widespread in French raps. In short, there is clearly a distinction in both the most important features and the language of raps.

Feature Importance (Feature Set)

This plot clearly shows which features are important and the information from it can clearly be used to get some idea about which features would possibly contribute most to, for example, clustering. However, the problem is that if the code is run over and over again each time, the top few features may vary more or less slightly. It can clearly be seen that features like speechiness, the 9th coefficient of the timbre vector, and instrumentalness are the most important features. There are a total of eight features whose value of “importance” is higher than 0.5. So I decided to use these features in the “Heatmap” section (left button). The most important two features: speechiness and the 9th coefficient of the timbre vector will be used in “Feature Importance (feature Set Plot)” (right button) to see how the tracks differ from each other and whether Dutch and French tracks clearly form a separate cluster on their own or not.

Feature Importance (Feature Set Plot)

In these plots, the features “speechiness” are plotted against the 9th coefficient of the timbre vector. Additionally, the size of the bubbles represents the strength of the energy. Unfortunately, there are few data points used in this plot, which makes generalization and drawing conclusions less reliable and accurate. However, we can clearly see that French rap has lower speechiness and higher c09, while Dutch rap has a lower average c09 but a slightly higher speechiness. It is also evident that the tracks form clusters on their own.